Nvidia's problem isn't demand. It's supply

Nvidia is the most valuable company in the world, with a market value around $5.5 trillion, and it still trades at a multiple that sits below several software companies with a fraction of its growth. The reason is the shape of its forecast.

The stock changes hands near $225, about 28.3 times its adjusted earnings over the past twelve months. On the earnings analysts expect for fiscal 2027, the same price is about 24 times. On fiscal 2028 estimates it is roughly 14.5 times. The price is identical in all three cases; what differs is the earnings that price is buying.

The unusual part of the guidance

On its most recent earnings call, management guided to roughly 70% revenue growth for fiscal 2028 — the first time the company has guided a full year that far ahead. What stood out was the reason it did not guide higher. Chief Financial Officer Colette Kress told investors that customer forecasts point to growth doubling, and that the company expects roughly 70% because it is supply-constrained.

That is a very different sentence from a demand warning. It shifts the analytical question from "will customers keep buying?" to "how many systems can actually be built?" Management expects supply to remain the bottleneck at least through the end of fiscal 2028 and has described memory pricing as extreme — a cost line that is guided to push gross margin into the 72% to 73% range.

Vera Rubin and the per-gigawatt math

The company's newest platform, Vera Rubin, began production shipments in August and is expected to be the fastest product ramp in its history, with purchase orders in from every major hyperscaler, AI cloud and system builder. Management frames the opportunity in an unusual unit: roughly $40 billion of revenue per gigawatt of data center capacity, against about $25 billion with the prior Blackwell generation.

Thinking in gigawatts rather than in chips explains why the AI trade has become an energy and construction trade as well. It also explains why a single delayed campus — like the New Mexico project where Oracle has invoked force majeure — now registers as market news.

The parts of the bull case that deserve scrutiny

Three items belong on any honest list:

• Circular demand. Nvidia has invested close to $50 billion in frontier AI labs that are also customers. Management argues the compute can be redeployed elsewhere if a lab stumbles, but the exposure is real.
• Estimate dispersion. Analyst estimates for fiscal 2028 earnings run from about $12.62 to $17.49 per share. That is a wide band for the year the forward multiple rests on.
• Margin direction. Gross margin is guided lower on memory costs, even after price increases the company has already pushed through.

A supply-constrained monopoly looks cheap on forward estimates and expensive on trailing ones. Which number you use is the entire argument.

What it means for the rest of the chip complex

When the leader cannot meet demand, the overflow tends to find other suppliers — which is why memory makers, networking vendors, foundries and power-component companies have been carried along. It also means the sector's risk is increasingly about logistics rather than sentiment: components, packaging capacity and grid connections, not enthusiasm for AI.

For investors, the durable version of this idea is to watch the constraint rather than the narrative. When memory pricing eases and delivery times shorten, the supply story changes — and so does the reason the stock's forward multiple looks low.